Supervised Wireless Communication: An Analytic Framework for Real-Time Model Inference in the 5G Core Network

The base for providing intelligent management is evolving towards Beyond 5G (B5G) and Sixth Generation (6G) networks. The increasing demand for data traffic, and the deployment of a significant number of network slices, create an essential need to improve the performance of resource utilization and allocation. Deployment strategies for real-time network optimization become challenging with the trends in heterogeneity and diversity. This work proposes the Fifth Generation (5G) wireless communication’s real-time prediction framework by analyzing the traffic of each Network Function (NF) in the Core Network (CN) architecture, simulated in a containerized infrastructure. Based on a varying range of hyperparameters, regressive training is conducted, and an optimal model is chosen for the inference phase through model tracking and registry support. During the real-time prediction stage, if the comparison results in a larger difference, a messaging system is implemented to notify a specific authority for further investigation. Finally, the experimental result shows the feasibility of this proposal to forecast with high accuracy.

I. The Need For Real-Time Model Inference

One of the design criteria for Beyond 5G(B5G) and Sixth Generation (6G) systems is the optimization and enhancement of the system performance due to the explosive growth of data traffic. Thus, the requirement of fulfilling peak data rates -delivering up to 20 Gbit/s-, decreased latency, and enhanced reliability is facilitated by upcoming Next Generations. With the fast growth in data, the need for providing a stable and reliable network becomes a decisive constraint, as well as, the utilization and availability of the resources influence, to enable an efficient system [1], [2].